VectorChat | Enabling Decentralized Solutions for AI

General Information

About

Crafting the Premier User Experience

Decentralization

Candy: Bites of Knowledge

Productivity

Entertainment

General Information

About

Toffee is a cutting-edge, character-based conversational AI platform built upon decentralized inference and retrieval-augmented generation, enabling an unparalleled user experience. Designed to facilitate both productivity and entertainment, the platform seamlessly adapts to fit the unique needs of each user.

At the heart of the Toffee platform are “Candies”—user-created packs of knowledge designed to dramatically improve the quality of character interactions while simplifying the character creation process. These candies consist of various forms of media, including text, audio, and video. See Candy: Bites of Knowledge.

Crafting the Premier User Experience

The platform is designed to feel intuitive and user-friendly, evoking a sense of familiarity even for first-time users. The one-page modular design of the site streamlines navigation while allowing for maximum customizability, letting users control what is shown or hide components altogether.

Toffee addresses common pain points in conversational AI. On leading platforms, users often encounter daunting, open-ended questions when creating characters, hindering accessibility and engagement. As a result, character creation on those platforms is unfriendly to most users and often results in lower quality characters.

Instead, throughout the platform, the ⟡ AI assistance icon is present wherever significant text input is needed. Clicking this icon pulls up a suite of AI tools to simplify the task of writing while ensuring correct grammar for the LLM. The icon also appears wherever images are needed, allowing every character or candy to have an image even if the author didn’t prepare one.

When combined with candies, users can create highly accurate and lifelike characters from just an idea, enhancing both accessibility and engagement.

Decentralization

Perhaps the most frustrating experience for users is being censored, a common occurrence in conversational AI. Toffee minimizes censorship, being particularly lenient within private chats and carefully tailoring recommendations to present controversial characters only to those open to such characters. By default, safe search is enabled.

Decentralization allows for the freedom present on the Toffee platform, enabled through uncensorable, permissionless inference. Gradually replacing in-house and third-party services with permissionless solutions, Toffee maximizes user freedom and minimizes restrictions.

Candy: Bites of Knowledge

The Magic of Candies

Candies are user-created “bites of knowledge” that enhance characters with extensive information and immersive instructions on mannerisms and speech. These candies enable realistic characters, large knowledge bases, and effectively infinite, meaningful conversations by allowing characters to intelligently access information beyond the context limit of an LLM.

Functionality

Users can create and share candies containing vast amounts of data, applying them to new and existing characters to personalize and enhance interactions. There is no limit to the number of candies that can be attached to a character, allowing for endless customization.

At launch, candies support text, images, and links to media on platforms like YouTube and Spotify. As the Toffee platform evolves, this will include other modalities such as video and audio.

Why are they needed?

Candies address common issues in conversational AI, such as context limitations and hallucination, through Retrieval-Augmented Generation (RAG). Instead of placing the entire conversation history and knowledge base into the context window of an LLM, Toffee intelligently includes only the most relevant information for each query, reducing the cost of inference and enabling larger knowledge bases. As Toffee advances the efficacy of RAG, interactions become increasingly relevant, accurate, and immersive. Learn More.

Productivity

Overview

Toffee enhances productivity in conversational AI through supporting extensive data and near-infinite conversation lengths, enabling the creation of powerful assistants and vast knowledge bases.

Multi-modality

Toffee allows users to incorporate a significant amount of data into conversations–in addition to the context provided through applied candies. Users can upload text files, PDFs, and images during conversations.

These files are stored in a way akin to that of a candy local to this specific instance of this character. Users can choose to keep them local or export them as a candy for use with other characters or in other instances with the same character.

Knowledge Base

Many productive tasks require immense context, such as detailed company information, project timelines, or relevant past work examples. Candies allow users to upload vast amounts of data, including documents, PDFs, and books, enhancing characters with relevant information. As the platform evolves, the growing collective knowledge shared by the community makes it easy to enhance characters without compiling data from scratch.

Export Valuable Context

Toffee offers a seamless method for preserving valuable context accrued during conversation. User conversation history is stored in private candies unique to that user and to the specific instance of the character in question. As a result, users are able to export their conversation history as a candy, allowing for the seamless transfer of rich context to other characters or additional instances of the same character.

A common example is writing, where the AI assistant gradually learns the user’s style, word choice, and niche information about the topic. When a new instance of the AI is needed, users can add this rich context as a candy, avoiding the need to rebuild it from scratch

Entertainment

Overview

Toffee delivers the ultimate conversational AI experience through endless, quality, and highly personalized interaction without restriction.

Immersive Conversation

Toffee strives to create an immersive experience, making interactions with characters feel genuinely real. Advanced Retrieval Augmented-Generation (RAG) ensures that characters do not forget messages as conversations lengthen or as more knowledge is added. Instead, they become further personalized over time as the character evolves with the user.

Characters recall memories remarkably similar to that of humans. Even when not directly referencing memories or a previous moment in conversation, characters respond with subtlety and nuance only possible with advanced long-term memory.

Personalization

Personalization is at the core of the Toffee experience, enabling users to tailor interactions to
their specific interests and needs. Users can customize both public and private characters using
a vast library of community-created candies or with their own custom-created candies.

Users can craft their own characters from scratch, embedding them with significant amounts of
relevant knowledge or with intricate instruction on how to communicate. Alternatively, they can
build upon the extensive knowledge base of candies created by the community, quickly
providing characters with rich context without having to compile data from scratch.

Fine-tuned customization is possible through the lore book feature, which allows users to imbue
characters with specific memories, instructions, or details that will be included in every LLM
query.

Personalization is at the core of the Toffee experience, enabling users to tailor interactions to their specific interests and needs. Users can customize both public and private characters using
a vast library of community-created candies or with their own custom-created candies.

Fine-tuned customization is possible through the lore book feature, which allows users to imbue
characters with specific memories, instructions, or details that will be included in every LLM
query.

Personalization is at the core of the Toffee experience, enabling users to tailor interactions to their specific interests and needs. Users can customize both public and private characters using a vast library of community-created candies or with their own custom-created candies.

Users can craft their own characters from scratch, embedding them with significant amounts of relevant knowledge or with intricate instruction on how to communicate. Alternatively, they can build upon the extensive knowledge base of candies created by the community, quickly providing characters with rich context without having to compile data from scratch.

Fine-tuned customization is possible through the lore book feature, which allows users to imbue characters with specific memories, instructions, or details that will be included in every LLM query.

Voice

Pushing the boundaries of conversational AI, Toffee delivers real-time, realistic audio
interactions, representing the next evolution in creating lifelike, dynamic conversations.

With just a 5-second audio clip, users can generate realistic voices, while a one-minute sample
allows for even higher fidelity. Characters communicate with voices that are authentic and
uniquely tailored to their identities. These generated voices can be kept private or shared with
the wider community, similar to characters and candies.

Users can call characters as they would a friend. By default, characters speak with voices
assigned by their creators; however, users can personalize the voice to one of their own or
select from voices created by Toffee or the community.

Pushing the boundaries of conversational AI, Toffee delivers real-time, realistic audio interactions, representing the next evolution in creating lifelike, dynamic conversations.

With just a 5-second audio clip, users can generate realistic voices, while a one-minute sample allows for even higher fidelity. Characters communicate with voices that are authentic and uniquely tailored to their identities. These generated voices can be kept private or shared with the wider community, similar to characters and candies.

Users can call characters as they would a friend. By default, characters speak with voices assigned by their creators; however, users can personalize the voice to one of their own or select from voices created by Toffee or the community.

Token

Chunking

Toffee

Company

Chunking

Try Toffee

Subscribe to the VectorChat newsletter